《计算机应用》唯一官方网站 ›› 0, Vol. ›› Issue (): 257-261.DOI: 10.11772/j.issn.1001-9081.2024020210

• 多媒体计算与计算机仿真 • 上一篇    下一篇

基于二维空间遍历长短时记忆网络的端到端车牌识别方法

王骥1, 黄远甲1, 方炜1, 谢文武2(), 熊文昌3, 李兴旺4   

  1. 1.华中师范大学 物理科学与技术学院,武汉 430079
    2.湖南理工学院 信息科学与工程学院,湖南 岳阳 414000
    3.上海脉衍人工智能科技有限公司,上海 200032
    4.河南理工大学 物理与电子信息学院,河南 焦作 454000
  • 收稿日期:2024-03-04 修回日期:2024-04-10 接受日期:2024-04-24 发布日期:2025-01-24 出版日期:2024-12-31
  • 通讯作者: 谢文武
  • 作者简介:王骥(1987—),男,湖北襄阳人,副教授,博士,主要研究方向:B5G/6G移动通信
    黄远甲(1996—),男,河南信阳人,硕士研究生,主要研究方向:图像处理、计算机视觉
    方炜(2000—),男,湖北孝感人,硕士研究生,主要研究方向:室内定位
    谢文武(1979—),男,湖北荆州人,副教授,博士,主要研究方向:无线通信系统算法、教育大数据
    熊文昌(1985—),男,江苏南京人,博士,主要研究方向:图像处理、计算机视觉
    李兴旺(1981—),河南焦作人,副教授,博士,主要研究方向:5G及6G物理层传输技术。
  • 基金资助:
    国家自然科学基金资助项目(62101205);湖北省重点研发计划项目(2023BAB061)

End-to-end license plate recognition method based on 2D space traversal LSTM network

Ji WANG1, Yuanjia HUANG1, Wei FANG1, Wenwu XIE2(), Wenchang XIONG3, Xingwang LI4   

  1. 1.College of Physical Science and Technology,Central China Normal University,Wuhan Hubei 430079,China
    2.School of Information Science and Engineering,Hunan University of Science and Technology,Yueyang Hunan 414000,China
    3.Shanghai Pulspread Artificial Intelligence Technology Company Limited,Shanghai 200032,China
    4.School of Physics and Electronic Information Engineering,Henan Polytechnic University,Jiaozuo Henan 454000,China
  • Received:2024-03-04 Revised:2024-04-10 Accepted:2024-04-24 Online:2025-01-24 Published:2024-12-31
  • Contact: Wenwu XIE

摘要:

车牌识别方法是现代智能交通管理系统的重要组成部分,在许多领域得到了广泛的应用然而,在实际应用场景中,存在多行牌照,传统方法在处理多行车牌时灵活性不够,无法实现高精度端到端识别。为此,提出一种基于二维空间遍历长短时记忆(2DST-LSTM)网络的端到端识别方法识别单行车牌和双行车牌。所提方法摒弃了以往的图像分割步骤,而以端到端的方式识别车牌,使车牌识别的效能和精度更高。2DST-LSTM可以提高车牌,尤其是双行车牌,在复杂环境下的识别效果。在多个数据集上的实验结果表明,所提方法对双行车牌的识别率最高达到了98.6%,证明了所提方法的有效性。

关键词: 车牌识别, 双行车牌, 端到端网络, 深度学习, 卷积神经网络

Abstract:

License plate recognition method is an important part of modern intelligent traffic management system and has been widely used in many fields. However, in practical application scenarios, there are multi-row license plates, and the traditional methods are not flexible enough to deal with multi-row license plates, and cannot achieve high-precision end-to-end recognition. Therefore, an end-to-end recognition method based on 2D Space Traversal Long Short-Term Memory (2DST-LSTM) network was proposed to recognize single-row and double-row license plates. The proposed method abandons the previous image segmentation step, and carries out license plate recognition in an end-to-end way, which makes the license plate recognition more efficient and accurate. 2DST-LSTM can improve the recognition effect of license plates, especially double-row license plates, in complex environment. Experimental results on multiple datasets show that the proposed method achieves the highest recognition rate up to 98.6% for double-row license plates, which verified its effectiveness.

Key words: license plate recognition, double-row license plate, end-to-end network, deep learning, convolutional neural network

中图分类号: